why pad S repeated numbers of the last value to the end of the original sequence before patching about patchtst

[linfeng-du]

Since S does not necessarily devide L-P, there could be unused values if we do not pad additional values. Padding S repeated numbers is the same as padding the sequence to the cloest number divisble by S since patch_num is only increased by 1 and is used to segment the patches via torch.unfold later on.

[yuqinie98]

Thanks @David-Ttao for asking and @linfeng-du for reply! The reason is that almost all the points are used twice in overlapped patching setting, but the last few points, which in principle are the most important, would be only used once if we do not pad it.

[David-Ttao]

Since S does not necessarily devide L-P, there could be unused values if we do not pad additional values. Padding S repeated numbers is the same as padding the sequence to the cloest number divisble by S since patch_num is only increased by 1 and is used to segment the patches via torch.unfold later on.

thx for your response, i have understanded it

[David-Ttao]

Thanks @David-Ttao for asking and @linfeng-du for reply! The reason is that almost all the points are used twice in overlapped patching setting, but the last few points, which in principle are the most important, would be only used once if we do not pad it.

thx for your response, i have understanded it!!